Miguel Rodriguez

Open-source multi-agent coordination framework: Claude, Gemini and open-weight models working as a coordinated fleet with shared state, cross-model peer review and operational safeguards.

Cyber-physical AI system combining a Huenit robot arm, model reasoning, locally hosted Apertus and a G-code controller. It also generates a living Automated Technical File from source code and operational logs.

On-device RAG platform for field and industrial teams. Cited answers from an organisation's own documents, fully offline via MLX and Apertus, with iOS and web interfaces.

On-device LLM interpretability: a forward-only disposition lens that reads model state from J-space and visualises confidence, uncertainty, refusal and evasiveness before the model speaks.

QLoRA fine-tune of Apertus 8B trained on operational data from the Flotilla fleet. A four-arm blind evaluation found that retrieval explained the observed performance gain; under identical retrieved context the adapter performed worse than the base model. Corpus, adapter and per-question scores are public.

On-device iOS oracle using a local vision-language model and Ministral 3B. Fully local inference via MLX.

Pipeline that turns manuscripts into short vertical microdramas using generative video, voice, music and automated editing.

AI-assisted music project reimagining classical works across Celtic, vallenato, progressive house, EDM and other genres.